Class CDOptimizeBase
- java.lang.Object
-
- gov.nih.mipav.view.renderer.WildMagic.AAM.CDOptimizeBase
-
- Direct Known Subclasses:
CDOptimizeBFGS
,CDOptimizeCG
,CDOptimizePS
,CDOptimizeSA
,CDOptimizeSD
public abstract class CDOptimizeBase extends java.lang.Object
This is the Java modified version of C++ active appearance model API (AAM_API). It is modified with a subset of required functions for automatic MRI prostate segmentation. AAM-API LICENSE - file: license.txt This software is freely available for non-commercial use such as research and education. Please see the full disclaimer below. All publications describing work using this software should cite the reference given below. Copyright (c) 2000-2003 Mikkel B. Stegmann, mbs@imm.dtu.dk IMM, Informatics & Mathematical Modelling DTU, Technical University of Denmark Richard Petersens Plads, Building 321 DK-2800 Lyngby, Denmark http://www.imm.dtu.dk/~aam/ REFERENCES Please use the reference below, when writing articles, reports etc. where the AAM-API has been used. A draft version the article is available from the homepage. I will be happy to receive pre- or reprints of such articles. /Mikkel ------------- M. B. Stegmann, B. K. Ersboll, R. Larsen, "FAME -- A Flexible Appearance Modelling Environment", IEEE Transactions on Medical Imaging, IEEE, 2003 (to appear) ------------- 3RD PART SOFTWARE The software is partly based on the following libraries: - The Microsoft(tm) Vision Software Developers Kit, VisSDK - LAPACK DISCLAIMER This software is provided 'as-is', without any express or implied warranty. In no event will the author be held liable for any damages arising from the use of this software. Permission is granted to anyone to use this software for any non-commercial purpose, and to alter it, subject to the following restrictions: 1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. 2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software. 3. This notice may not be removed or altered from any source distribution. -- No guarantees of performance accompany this software, nor is any responsibility assumed on the part of the author or IMM. This software is provided by Mikkel B. Stegmann and IMM ``as is'' and any express or implied warranties, including, but not limited to, the implied warranties of merchantability and fitness for a particular purpose are disclaimed. In no event shall IMM or Mikkel B. Stegmann be liable for any direct, indirect, incidental, special, exemplary, or consequential damages (including, but not limited to, procurement of substitute goods or services; loss of use, data, or profits; or business interruption) however caused and on any theory of liability, whether in contract, strict liability, or tort (including negligence or otherwise) arising in any way out of the use of this software, even if advised of the possibility of such damage. $Revision: 1.4 $ $Date: 2003/04/23 14:49:15 $ Optimization base class. This abstract class is the base class for all classes implementing specific optimization procedures. author: Rune Fisker, 26/1-1999- Author:
- Ruida Cheng
-
-
Field Summary
Fields Modifier and Type Field Description static int
CentralDifference
int
ENumGrad
static int
eoptBFGS
static int
eoptConjugateGradient
int
EOptMethod
static int
eoptPatternSearch
static int
eoptSimulatedAnnealing
static int
eoptSteepestDescent
static int
eoptUnknown
int
ETermCode
static int
etermConsecMaxStepMax
static int
etermDeltaFuncVal
static int
etermGradTol
static int
etermLineSearch
static int
etermMaxFuncEval
static int
etermMaxIterations
static int
etermNoStop
static int
etermStepTol
static int
etermUnknown
static int
FitLine
static int
ForwardDifference
double
m_dDeltaFuncVal
value for stop criterion: abs(f - fplus) < m_dDeltaFuncValdouble
m_dEta
machine precision.double
m_dGradTol
A positive scalar giving the tolerance at which the scaled gradient in considered close enough to zero to terminate the algorithmdouble
m_dMachEps
Machine precisiondouble
m_dMaxStep
A positive scalar giving the maximum allowable scaled steplength at any iteration. maxstep is used to prevent steps that would cause the optimazation algorithm to overflow or leave the domain of interest, as well as to detect divergence.double
m_dStepTol
A positive scalar giving the tolerance at which the scaled distance between two successive iterated is considered close enough to zero to terminate the algorithmdouble
m_dTypF
positive scalar estimating the magnitude of f(x) near he minimizer x-star.int
m_eNumGrad
methods for estimating the gradient nummericalyboolean
m_fAnalyticGrad
flag indicating if the gradient is to be calc. analyticaly or nummericalyboolean
m_fLogFuncValues
Logical variable which determines whether function parameters and corresponding return values should be stored for later analysis.int
m_iStopCriteria
holds the stop criteria in use, e.g m_iStopCriteria = etermMaxFuncEval | etermMaxIterationsint
m_nConsecMax
Number of conseccutive past steps whose scaled length was equal to maxstepint
m_nConsecMaxStepMax
max number of conseccutive past steps whose scaled length was equal to maxstep to terminateint
m_nFDigits
A positive integer specifying the number of reliable digits returned by the objective function FN. fdigits is used to set the parameter n (eta) that is used in the code to specify the relative noise in f(x); the main use of eta is in calculation finite difference step size. eta is set to macheps if fdigits = -1.int
m_nFuncEval
counter to the number of function evaluations.int
m_nGradEval
counter to the number of gradient evaluations.int
m_nIterations
counter to the number of iterations.private int
m_nMaxFuncEval
limit for number of function evaluations.private int
m_nMaxIterations
A positive integer specifying the maximum number of iterations that may be performed before the algorithm is halted.private CDOptimizeFuncBase
m_pFuncEvalBase
pointer to the function to be minimized.CDVector
m_vFuncVal
private CDVector
m_vMethodPar
special parameters used by each optimization method e.g. step size used for calc. nummerical gradientCDVector
m_vNFuncEval
java.util.Vector<CDVector>
m_vvFuncParm
-
Constructor Summary
Constructors Constructor Description CDOptimizeBase()
Constructor
-
Method Summary
All Methods Instance Methods Abstract Methods Concrete Methods Modifier and Type Method Description double
EvalFunction(CDVector x)
function and gradient evaluation methodsvoid
EvalGradient(CDVector x, CDVector gc, double dFuncVal)
function and gradient evaluation methodsint
ExactLineSearch(CDVector xc, double fc, CDVector g, CDVector p, CDVector xplus, double[] fplus, boolean[] maxtaken)
Perform exact line search Input: xc: parameter fc: function value at xc g: gradient at xc p: search direction Output: xplus: new parameter fplus: function value for new parameter maxtaken: max taken in line search return termination codeCDOptimizeFuncBase
GetFuncEvalBase()
get point to func eval. base.int
LineSearch(CDVector xc, double fc, CDVector g, CDVector p, CDVector xplus, double[] fplus, boolean[] maxtaken)
Warpper for line searchint
LineSearch(CDVector xc, double fc, CDVector g, CDVector p, CDVector xplus, double[] fplus, boolean[] maxtaken, boolean fSoft)
int
MaxFuncEval()
get the number of evaluationsint
MaxIterations()
get limit for the number of iterationsCDVector
MethodPar()
get numerical gradient functionabstract int
Minimize(CDVector x, CDOptimizeFuncBase pFuncEvalBase)
the Minimize function using analytic gradientint
MinimizeNum(CDVector x, CDOptimizeFuncBase pFuncEvalBase, CDVector vMethodPar)
java.lang.String
Name()
name of optimization methodevoid
NumGrad(CDVector x, CDVector gradient, double dFuncVal)
calculate nummerical gradient function usingdecided gradient calculation method input: x: parameter dFuncVal: function value in x output: gradient: gradientint
OptMethod()
name of optimization methodvoid
SetFuncEvalBase(CDOptimizeFuncBase pFuncEval)
set point to func eval. base.void
SetMachineEps()
----------------------------[ MachineEps ]---------------------------- Calculate machine epsilon Algorithm A1.3.1 - p. 303 Dennis and Schnabel, Numerical Methods for Unconstrained Optimization and Nonlinear Equations 1983, Prentice-Hallvoid
SetMaxFuncEval(int nMaxFuncEval)
set limit for the number of function evaluationsvoid
SetMaxIterations(int nMaxIterations)
set limit for the number of iterationsvoid
SetMethodPar(CDVector vMethodPar)
set numerical gradient functionint
SoftLineSearch(CDVector xc, double fc, CDVector g, CDVector sn, CDVector xplus, double[] fplus, boolean[] maxtaken)
Perform line search Given g'p < 0 and alpha < 1/2 (alpha = 1e-4 is used), find plus = xc + lambda p,lambda in [0;1], such that f(xplus) <= f(xc) + alpha * lambda * g'p, using backtracking line search Algorithm A6.3.1 p. 325 Dennis and Schnabel, Numerical Methods for Unconstrained Optimization and Nonlinear Equations 1983, Prentice-Hall Input: xc: parameter fc: function value at xc g: gradient at xc sn: search direction Output: xplus: new parameter fplus: function value for new parameter maxtaken: max taken in line search return termination codeint
UmStop(CDVector x, CDVector xplus, double f, double fplus, CDVector g, int retcode, boolean maxtaken)
Decide wether to terminate minimization Modified version of Algorithm A7.2.1 p. 347 Dennis and Schnabel, Numerical Methods for Unconstrained Optimization and Nonlinear Equations 1983, Prentice-Hall Input: x: parameter xplus: new parameter fplus: function value for new parameter g: gradient at x retcode: return code from line search maxtaken: max taken in line search Output: return termination codeint
UmStop0(CDVector x0, double functionValue, CDVector gradient)
-
-
-
Field Detail
-
ENumGrad
public int ENumGrad
-
ForwardDifference
public static int ForwardDifference
-
CentralDifference
public static int CentralDifference
-
FitLine
public static int FitLine
-
EOptMethod
public int EOptMethod
-
eoptUnknown
public static int eoptUnknown
-
eoptSteepestDescent
public static int eoptSteepestDescent
-
eoptConjugateGradient
public static int eoptConjugateGradient
-
eoptBFGS
public static int eoptBFGS
-
eoptSimulatedAnnealing
public static int eoptSimulatedAnnealing
-
eoptPatternSearch
public static int eoptPatternSearch
-
ETermCode
public int ETermCode
-
etermNoStop
public static int etermNoStop
-
etermGradTol
public static int etermGradTol
-
etermStepTol
public static int etermStepTol
-
etermLineSearch
public static int etermLineSearch
-
etermConsecMaxStepMax
public static int etermConsecMaxStepMax
-
etermDeltaFuncVal
public static int etermDeltaFuncVal
-
etermMaxFuncEval
public static int etermMaxFuncEval
-
etermMaxIterations
public static int etermMaxIterations
-
etermUnknown
public static int etermUnknown
-
m_dMachEps
public double m_dMachEps
Machine precision
-
m_dTypF
public double m_dTypF
positive scalar estimating the magnitude of f(x) near he minimizer x-star. It is only used in the gradient stopping condition given below. typf should be approximately |f(x*)|
-
m_nFDigits
public int m_nFDigits
A positive integer specifying the number of reliable digits returned by the objective function FN. fdigits is used to set the parameter n (eta) that is used in the code to specify the relative noise in f(x); the main use of eta is in calculation finite difference step size. eta is set to macheps if fdigits = -1. If f(x) is suspected to be noisy but the approximate value of fdigits is unknown, it should be estimated be the routine of Hamming[1973] given in Gill, Murray and Wright[1981]
-
m_dMaxStep
public double m_dMaxStep
A positive scalar giving the maximum allowable scaled steplength at any iteration. maxstep is used to prevent steps that would cause the optimazation algorithm to overflow or leave the domain of interest, as well as to detect divergence. It should be chosen small enough to prevent the first two of these occurrences but larger than any anticipated reasonable stepsize. The algorithm will halt if it takes steps of length maxstep on m_nConsecMaxStepMax conseccutive iterations
-
m_nConsecMax
public int m_nConsecMax
Number of conseccutive past steps whose scaled length was equal to maxstep
-
m_dEta
public double m_dEta
machine precision.
-
m_fAnalyticGrad
public boolean m_fAnalyticGrad
flag indicating if the gradient is to be calc. analyticaly or nummericaly
-
m_eNumGrad
public int m_eNumGrad
methods for estimating the gradient nummericaly
-
m_iStopCriteria
public int m_iStopCriteria
holds the stop criteria in use, e.g m_iStopCriteria = etermMaxFuncEval | etermMaxIterations
-
m_dGradTol
public double m_dGradTol
A positive scalar giving the tolerance at which the scaled gradient in considered close enough to zero to terminate the algorithm
-
m_dStepTol
public double m_dStepTol
A positive scalar giving the tolerance at which the scaled distance between two successive iterated is considered close enough to zero to terminate the algorithm
-
m_nConsecMaxStepMax
public int m_nConsecMaxStepMax
max number of conseccutive past steps whose scaled length was equal to maxstep to terminate
-
m_dDeltaFuncVal
public double m_dDeltaFuncVal
value for stop criterion: abs(f - fplus) < m_dDeltaFuncVal
-
m_nIterations
public int m_nIterations
counter to the number of iterations.
-
m_nFuncEval
public int m_nFuncEval
counter to the number of function evaluations.
-
m_nGradEval
public int m_nGradEval
counter to the number of gradient evaluations.
-
m_fLogFuncValues
public boolean m_fLogFuncValues
Logical variable which determines whether function parameters and corresponding return values should be stored for later analysis.
-
m_vFuncVal
public CDVector m_vFuncVal
-
m_vNFuncEval
public CDVector m_vNFuncEval
-
m_vvFuncParm
public java.util.Vector<CDVector> m_vvFuncParm
-
m_pFuncEvalBase
private CDOptimizeFuncBase m_pFuncEvalBase
pointer to the function to be minimized.
-
m_vMethodPar
private CDVector m_vMethodPar
special parameters used by each optimization method e.g. step size used for calc. nummerical gradient
-
m_nMaxIterations
private int m_nMaxIterations
A positive integer specifying the maximum number of iterations that may be performed before the algorithm is halted. Appropriate values depend strongly on the dimension and difficulty of the problem, and the cost of evaluating the nonlinear function.
-
m_nMaxFuncEval
private int m_nMaxFuncEval
limit for number of function evaluations.
-
-
Method Detail
-
Minimize
public abstract int Minimize(CDVector x, CDOptimizeFuncBase pFuncEvalBase)
the Minimize function using analytic gradient- Parameters:
x
-pFuncEvalBase
-- Returns:
-
Name
public java.lang.String Name()
name of optimization methode- Returns:
-
OptMethod
public int OptMethod()
name of optimization method
-
EvalFunction
public double EvalFunction(CDVector x)
function and gradient evaluation methods- Parameters:
x
-- Returns:
-
EvalGradient
public void EvalGradient(CDVector x, CDVector gc, double dFuncVal)
function and gradient evaluation methods- Parameters:
x
-gc
-dFuncVal
-
-
MethodPar
public CDVector MethodPar()
get numerical gradient function- Returns:
-
SetMethodPar
public void SetMethodPar(CDVector vMethodPar)
set numerical gradient function- Parameters:
vMethodPar
-
-
MaxIterations
public int MaxIterations()
get limit for the number of iterations- Returns:
-
MaxFuncEval
public int MaxFuncEval()
get the number of evaluations- Returns:
-
SetFuncEvalBase
public void SetFuncEvalBase(CDOptimizeFuncBase pFuncEval)
set point to func eval. base.- Parameters:
pFuncEval
-
-
GetFuncEvalBase
public CDOptimizeFuncBase GetFuncEvalBase()
get point to func eval. base.- Returns:
-
SetMaxIterations
public void SetMaxIterations(int nMaxIterations)
set limit for the number of iterations- Parameters:
nMaxIterations
- max iterations
-
SetMaxFuncEval
public void SetMaxFuncEval(int nMaxFuncEval)
set limit for the number of function evaluations- Parameters:
nMaxFuncEval
- max number of evaluations.
-
SetMachineEps
public void SetMachineEps()
----------------------------[ MachineEps ]---------------------------- Calculate machine epsilon Algorithm A1.3.1 - p. 303 Dennis and Schnabel, Numerical Methods for Unconstrained Optimization and Nonlinear Equations 1983, Prentice-Hall
-
UmStop
public int UmStop(CDVector x, CDVector xplus, double f, double fplus, CDVector g, int retcode, boolean maxtaken)
Decide wether to terminate minimization Modified version of Algorithm A7.2.1 p. 347 Dennis and Schnabel, Numerical Methods for Unconstrained Optimization and Nonlinear Equations 1983, Prentice-Hall Input: x: parameter xplus: new parameter fplus: function value for new parameter g: gradient at x retcode: return code from line search maxtaken: max taken in line search Output: return termination code
-
LineSearch
public int LineSearch(CDVector xc, double fc, CDVector g, CDVector p, CDVector xplus, double[] fplus, boolean[] maxtaken)
Warpper for line search- Parameters:
xc
- parameterfc
- function value at xcg
- gradient at xcp
- search directionxplus
- new parameterfplus
- function value for new parametermaxtaken
- max taken in line search- Returns:
- termination code
-
LineSearch
public int LineSearch(CDVector xc, double fc, CDVector g, CDVector p, CDVector xplus, double[] fplus, boolean[] maxtaken, boolean fSoft)
-
ExactLineSearch
public int ExactLineSearch(CDVector xc, double fc, CDVector g, CDVector p, CDVector xplus, double[] fplus, boolean[] maxtaken)
Perform exact line search Input: xc: parameter fc: function value at xc g: gradient at xc p: search direction Output: xplus: new parameter fplus: function value for new parameter maxtaken: max taken in line search return termination code
-
SoftLineSearch
public int SoftLineSearch(CDVector xc, double fc, CDVector g, CDVector sn, CDVector xplus, double[] fplus, boolean[] maxtaken)
Perform line search Given g'p < 0 and alpha < 1/2 (alpha = 1e-4 is used), find plus = xc + lambda p,lambda in [0;1], such that f(xplus) <= f(xc) + alpha * lambda * g'p, using backtracking line search Algorithm A6.3.1 p. 325 Dennis and Schnabel, Numerical Methods for Unconstrained Optimization and Nonlinear Equations 1983, Prentice-Hall Input: xc: parameter fc: function value at xc g: gradient at xc sn: search direction Output: xplus: new parameter fplus: function value for new parameter maxtaken: max taken in line search return termination code
-
NumGrad
public void NumGrad(CDVector x, CDVector gradient, double dFuncVal)
calculate nummerical gradient function usingdecided gradient calculation method input: x: parameter dFuncVal: function value in x output: gradient: gradient
-
MinimizeNum
public int MinimizeNum(CDVector x, CDOptimizeFuncBase pFuncEvalBase, CDVector vMethodPar)
-
-